Hybrid deep learning CNN-LSTM model for forecasting direct normal irradiance: a study on solar potential in Ghardaia,...

Area of Science:

• Renewable Energy Systems
• Machine Learning Applications
• Atmospheric Science

Background:

• Accurate solar radiance (SR) prediction is crucial for optimizing solar energy systems and understanding atmospheric phenomena.
• Existing machine learning models for SR forecasting have limitations in accuracy and reliability for short-to-medium term predictions.
• The integration of deep learning techniques offers potential for enhanced SR prediction capabilities.

Purpose of the Study:

• To conduct an in-depth analysis and performance evaluation of four distinct Solar Radiance (SR) prediction models.
• To compare the efficacy of Feed-forward Back Propagation (FFBP), Convolutional Feed-forward Back Propagation (CFBP), Support Vector Regression (SVR), and a hybrid Convolutional Neural Networks-Long Short-Term Memory (CNN-LSTM) deep learning model.
• To assess model performance using statistical indicators such as Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Normalized Root Mean Squared Error (nRMSE).

Main Methods:

• Implementation of four machine learning algorithms: FFBP, CFBP, SVR, and a hybrid CNN-LSTM deep learning model.
• Training and validation of models using historical experimental data for SR prediction.
• Quantitative evaluation of model predictions against actual recorded data using established statistical error metrics.

Main Results:

• The hybrid CNN-LSTM model consistently outperformed FFBP, CFBP, and SVR models in SR prediction accuracy and reliability.
• The CNN-LSTM model exhibited the lowest error rates across all statistical indicators used for evaluation.
• A high coefficient of determination (R² = 0.99925) was achieved by the CNN-LSTM model, indicating excellent predictive power.

Conclusions:

• The hybrid CNN-LSTM deep learning model represents a significant advancement in solar radiance prediction technology.
• This advanced model offers enhanced accuracy and reliability for forecasting solar radiance from a few hours to several days ahead.
• The findings suggest the CNN-LSTM model is a highly effective tool for applications requiring precise solar radiance forecasts, such as in renewable energy management.